A Low-Voltage Distribution Area Substation Relationship Identification Method Based on Voltage-Active Power Response Separation

By constructing a voltage-active power response separation model and an alternating minimization method, the accuracy and robustness of identifying the relationship between households and transformers in low-voltage distribution areas under the increased penetration rate of distributed photovoltaics were solved. This enabled effective handling of photovoltaic disturbances and data acquisition anomalies, and improved the accuracy and interpretability of household-transformer relationship identification.

CN123121485APending Publication Date: 2026-11-17CHINA THREE GORGES UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611228633.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-13
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

With the increasing penetration rate of distributed photovoltaic (PV) power, existing methods for identifying the relationship between low-voltage transformer substations and households are affected by PV disturbances and abnormal data collection, leading to a decrease in identification accuracy. It is difficult to accurately differentiate the contributions of the common voltage of the distributed transformer, line voltage drop, and local PV uplift. Furthermore, outliers and missing values ​​in the field-collected data interfere with the robustness and interpretability of the identification.

Method used

By constructing a voltage-active power response separation model, the local positive voltage correction component caused by photovoltaic self-consumption is independently separated. The data quality weight and comprehensive matching distance criterion are integrated to identify the period of photovoltaic self-consumption impact. An optimization model is established to identify the relationship between households and transformers. The parameters are solved by the alternating minimization method.

Benefits of technology

It effectively reduces the interference of photovoltaic disturbances on the identification of household transformer relationships, accurately identifies the period of photovoltaic self-consumption impact, enhances the robustness to anomalies in collected data, and improves the accuracy and interpretability of identification. It is suitable for online or offline identification tasks in large-scale low-voltage distribution areas.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123121485A_ABST
    Figure CN123121485A_ABST
Patent Text Reader

Abstract

This invention presents a low-voltage transformer-household relationship identification method based on voltage and active power response separation. First, it identifies outliers, completes missing values, and assigns data quality weights to the user voltage, active power, and low-voltage side voltage sequences of candidate transformers. Then, it identifies the photovoltaic (PV) self-consumption impact period using the joint characteristics of positive voltage rise and active power fall, constructing a PV impact indicator matrix. For user candidate transformer pairs, it constructs a voltage and active power response separation model, decomposing the user voltage into candidate transformer common voltage, long-term static voltage bias, etc. It estimates the parameters of each component using a constrained alternating minimization method, calculates the relevant distance and normalized reconstruction error after deducting local corrections, and fuses them to construct a comprehensive matching distance. Based on the optimal transformer matching distance and its distinguishability with the second-best transformer, it determines the transformer affiliation and outputs a list for users with low confidence. This invention can mitigate the impact of distributed PV disturbances and data acquisition anomalies, achieving robust and interpretable transformer-household relationship identification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of low-voltage distribution network topology identification and intelligent distribution operation and maintenance technology, specifically to a method for identifying the relationship between users and distribution transformers in low-voltage distribution areas in response to distributed photovoltaic disturbances. In particular, it relates to a method for identifying the affiliation relationship between users and distribution transformers through data preprocessing, identification of the period of impact of user-side photovoltaic self-consumption, voltage-active power response separation, constrained alternating minimization solution, and comprehensive matching score of candidate distribution transformers. Background Technology

[0002] The relationship between a low-voltage distribution area and its user transformer refers to the topological affiliation between a low-voltage user and its actual power supply transformer. The accuracy of this relationship directly impacts distribution area line loss calculation, low-voltage mitigation, three-phase imbalance mitigation, fault location, power supply reliability assessment, anti-theft analysis, and the construction of digital distribution network archives. With the widespread adoption of smart meters, distribution transformer monitoring terminals, and electricity consumption information collection systems, identifying the user transformer relationship based on the similarity between the user's voltage time series and the low-voltage side voltage time series of the distribution transformer has become the mainstream technical approach, independent of large-scale power outage tests and manual on-site inspections.

[0003] In traditional low-voltage distribution networks, users within the same transformer's power supply area share the same bus voltage and are affected by load fluctuations from the same source via similar line impedances. Consequently, the voltage curves of each user node exhibit a high degree of synchronicity in their temporal changes. Based on this physical characteristic, existing methods often employ similarity measures such as Pearson correlation coefficient, Euclidean distance, and dynamic time warping distance, combined with algorithms like hierarchical clustering, K-means clustering, or density clustering, to group users with similar voltage curve shapes to the same transformer, thereby achieving user-transformer relationship identification.

[0004] For example, in reference [1] Chinese patent CN110070089A, invention title: Low-voltage distribution area household transformer relationship identification method based on voltage curve similarity, by calculating the Pearson correlation coefficient between the user voltage sequence and the low-voltage side voltage sequence of each distribution transformer, the distribution transformer with the greatest correlation is determined as the belonging distribution transformer.

[0005] However, in recent years, the penetration rate of distributed photovoltaic (PV) power in low-voltage distribution areas has been continuously increasing, and the formation mechanism of user voltage curves has changed significantly. During the daytime self-consumption period of PV power generation, the active power absorbed by users from the grid decreases significantly or even reverses, resulting in a reduction in line voltage drop and a relative increase in user-side voltage. At the same time, factors such as inverter control strategies, module orientation, local shading, and cloud movement cause significant differences in the PV output characteristics of each user, resulting in discretization of voltage curves for users in the same distribution area due to differences in PV power. Users in different distribution areas may also be misclassified due to similar daytime PV voltage rise patterns.

[0006] Reference [2] Chinese Patent CN114021430A, Invention Title: A Method for Identifying the Relationship between Household Transformers in Low-Voltage Distribution Areas under Distributed Photovoltaic Access, which identifies and removes photovoltaic output time period data and uses the remaining data for similarity determination. Although this method weakens the impact of photovoltaic disturbances to a certain extent, simply cutting off photovoltaic time periods will lose effective information and is difficult to adapt to scenarios where only some time periods are used for photovoltaic self-consumption. It cannot finely distinguish the respective contributions of distribution transformer common voltage, line voltage drop and local photovoltaic rise.

[0007] Furthermore, the data collected on-site suffers from issues such as incomplete timestamp synchronization, isolated spikes, communication gaps, out-of-limit records, and metering errors. If the original voltage curves are directly used for similarity calculations, outliers and missing values ​​will severely interfere with the accurate estimation of correlation coefficients and reconstruction errors, reducing the robustness and interpretability of household-transformer relationship identification. Therefore, there is an urgent need for a method to simultaneously handle data acquisition anomalies, accurately identify the periods of impact from user-side photovoltaic self-consumption, and explicitly separate active power voltage drop and local photovoltaic positive voltage correction components for household-transformer relationship identification. Summary of the Invention

[0008] The technical problem to be solved by this invention is to provide a method for identifying the relationship between household transformers in low-voltage distribution areas based on voltage-active power response separation. By constructing a voltage-active power response separation model, the local positive voltage correction component caused by photovoltaic self-consumption is independently separated, and data quality weight and comprehensive matching distance criterion are integrated. This effectively weakens the impact of distributed photovoltaic disturbances and acquisition anomalies on the identification of household transformer relationships, and achieves robust and interpretable identification of household transformer relationships in low-voltage distribution areas.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: Based on voltage The method for identifying the relationship between low-voltage transformer substations and their customers based on active power response separation includes the following: S1: Obtain the user voltage sequence, user active power sequence and candidate distribution transformer low-voltage side voltage sequence within the target low-voltage distribution area, identify outliers, fill in missing values ​​and assign data quality weights to obtain preprocessed user voltage, user active power and distribution transformer voltage data. S2: Based on the positive rise of user voltage relative to the reference baseline and the decrease of user active power relative to the reference baseline after preprocessing, joint features are constructed to identify the periods of photovoltaic self-consumption impact, forming user... Momentary photovoltaic impact indicator matrix; S3: Build voltage for each user and each candidate transformer. The active power response separation model expresses the preprocessed user voltage as the sum of candidate distribution transformer common voltage, long-term static voltage bias, equivalent active voltage drop, local positive voltage correction component, and random error term; S4: Establish an optimization model with the sum of the weighted reconstruction error and the sparse penalty term of the local positive voltage correction component as the objective function; adopt the alternating minimization method to alternately estimate the long-term static voltage bias, active response coefficient and local positive voltage correction component under the conditions of satisfying the static voltage bias constraint and the active response coefficient constraint. S5: After deducting the local positive voltage correction component, the relevant distance and normalized reconstruction error are calculated based on the user disturbance removal voltage and the candidate distribution transformer common voltage, and then integrated to construct the comprehensive voltage matching distance. S6: For each user, candidate transformers are sorted according to the comprehensive voltage matching distance. The transformer to which the user belongs is determined based on the matching distance of the best candidate transformer and the difference in matching distance between the best and second-best candidate transformers.

[0010] Voltage in S3 The active power response separation model and its constraints are as follows: (9); (10); In the formula: This is the pre-processed user voltage; For candidate distribution transformers The pre-processed low-voltage side voltage is used as the candidate distribution transformer common voltage; The long-term static voltage bias to be estimated; The equivalent active voltage response coefficient to be estimated is denoted as . This refers to the preprocessed active power of the user. The local positive voltage correction component to be estimated; This is the random error term; This is the maximum permissible static voltage bias amplitude; This represents the upper limit of the active response coefficient.

[0011] The expressions for the objective function of the optimization model in S4 and the corresponding local positive correction penalty weights are as follows: (11); (12); In the formula: For users In candidate distribution transformers The objective function value is below. For the long-term static voltage bias to be estimated, Let be the equivalent active voltage response coefficient to be estimated. The local positive voltage correction component to be estimated; For data quality weights, For random error term, The total number of sampling points. The penalty weights are adjusted locally for positive correction. The local positive voltage correction component to be estimated. The overall sparsity coefficient is . To account for the disturbance penalty coefficient during the period of photovoltaic self-consumption, This is the disturbance penalty coefficient corresponding to the normal time period, and it satisfies... , For users Moment-time photovoltaic impact indicator matrix.

[0012] The alternating minimization method in S4 includes: S4.1: Let Given the current iteration number, and the local positive voltage correction component from the previous iteration. Calculate the equivalent voltage difference : (13); In the formula: For the first The equivalent voltage difference in the next iteration. For the preprocessed user voltage, For candidate distribution transformers Pre-processing low-voltage side output voltage, This indicates the local correction term from the previous round; S4.2: Solve a two-dimensional weighted least squares problem with interval constraints to update the long-term static voltage bias. and active response coefficient : (14); (15); (16); In the formula: For long-term static voltage bias. The active response coefficient, For the first The weighted sum of squared residuals in the next iteration The set of parameter values ​​that satisfy the constraints. b This represents the temporary long-term static voltage bias to be estimated. r This represents the temporary active response coefficient to be estimated. For data quality weights, For the first The equivalent voltage difference in the next iteration. This refers to the preprocessed active power of the user. This is the maximum permissible static voltage bias amplitude; This represents the upper limit of the active response coefficient.

[0013] S4.3: Calculate the positive residual based on the updated parameters. The local positive voltage correction component is updated using a non-negative soft threshold method. : (17); (18); In the formula: It is a positive surplus. For the preprocessed user voltage, For candidate distribution transformers Pre-processing low-voltage side output voltage, For long-term static voltage bias. The active response coefficient, This refers to the preprocessed active power of the user. This is a local positive voltage correction component. The penalty weights are adjusted locally for positive correction. For data quality weights, To prevent extremely small positive numbers with a denominator of zero; The retention threshold for the local positive voltage correction component; S4.4: Repeat steps S4.1 to S4.3 until the convergence condition is met. (19); In the formula: and The first Second and third The objective function value of the next iteration. This is the preset convergence threshold.

[0014] S5 includes: S5.1: Calculate the undisturbed voltage sequence Common voltage sequence with candidate distribution transformers The relevant distance between : (20); In the formula: For the relevant distance, The Pearson correlation coefficient; The elements are composed of constitute, For users At any moment Relative to candidate transformers The de-disturbance voltage; This is the pre-processed user voltage; This is a local positive voltage correction component; S5.2: Calculate the normalized reconstruction error, specifically: (twenty one); (twenty two); In the formula: For weighted root mean square voltage reconstruction error, For data quality weights, For random error term, To normalize the voltage reconstruction error, For users The preprocessed voltage sequence, The standard deviation operator; To prevent extremely small positive numbers with a denominator of zero.

[0015] S5.3: Integrate relevant distances and normalized reconstruction errors to construct a comprehensive voltage matching distance. : (twenty three); In the formula: For comprehensive voltage matching distance, The weights for the error term are used to adjust the contribution ratio of reconstruction error and curve shape distance to the matching distance. To normalize the voltage reconstruction error, For the perturbation voltage sequence Common voltage sequence with candidate distribution transformers The relevant distance between them.

[0016] The process of determining the user's assigned transformer in S6 is as follows: User All candidate distribution transformers are classified according to their comprehensive voltage matching distance. Sort by size from smallest to largest. and Calculate the discrimination index for the best and second-best candidate transformers. : (twenty four); In the formula: For the optimal candidate transformer distance Distance from the second-best candidate transformer difference; The rules for determining user ownership are as follows: (25); In the formula: Indicates user The final identification and attribution of the distribution transformer, This is the optimal candidate transformer. The optimal candidate transformer distance. To the maximum allowed matching distance, This is the minimum discrimination threshold.

[0017] This invention provides a method for identifying the relationship between low-voltage transformer substations and households based on voltage-active power response separation, which has the following technical advantages: 1) Effectively reduces the interference of distributed photovoltaic disturbances on the identification of household transformer relationships: This invention constructs a voltage... The active power response separation model independently separates the distribution transformer common voltage, long-term static voltage bias, equivalent active voltage drop, and local photovoltaic positive correction component in the user voltage. It decouples the local voltage rise caused by photovoltaic self-consumption from the mechanism level, avoiding the problem of users in the same area being misclassified due to photovoltaic differences, or users in different areas being misjudged due to similar photovoltaic forms. It significantly improves the identification accuracy of areas with a high proportion of distributed photovoltaic.

[0018] 2) Accurately identify the impact period of photovoltaic self-consumption on the user side and avoid information loss: This invention adopts the combined characteristics of positive voltage rise and active power decline, and combines adaptive threshold criteria to identify the impact period of photovoltaic self-consumption. Compared with the crude method of simply relying on voltage threshold or directly cutting off the full daytime data, it can more accurately locate the photovoltaic disturbance time, which not only retains the usable information of non-photovoltaic periods, but also provides a reliable basis for the time-sharing penalty of subsequent local correction terms.

[0019] 3) Enhanced robustness to anomalies in field-collected data: In the preprocessing stage, this invention identifies and completes anomalies and missing points, and assigns differentiated data quality weights to original valid sampling points, interpolation completion points, and anomaly correction points based on data reliability. This enables subsequent model parameter estimation and household-change relationship determination to automatically reduce the impact of low-quality data and improve adaptability to engineering conditions such as measurement noise, communication loss, and file errors.

[0020] 4) Improve the interpretability of the user-transformer relationship determination results: This invention removes disturbances from user voltage curves and constructs a unified matching distance by integrating relevant distances and normalized reconstruction errors. At the same time, it introduces a distinguishing criterion between the best candidate transformer and the second-best candidate transformer. This not only provides a physical modeling-based attribution conclusion, but also outputs on-site verification marks and candidate transformer lists for boundary users and suspected multi-attribution users, providing practical and effective reference for maintenance personnel to review and investigate.

[0021] 5) It has high computational adaptability and engineering applicability: The response separation model of each user and candidate distribution transformer pair in this invention can be solved independently. The overall computational complexity is linearly related to the number of users, the number of candidate distribution transformers and the number of sampling points. It is suitable for online or offline batch identification tasks of large-scale low-voltage distribution areas. At the same time, the alternating minimization iterative solution method adopted has stable convergence and can quickly obtain globally reasonable parameter estimation results.

[0022] 6) Improve the anti-interference capability of household transformer relationship determination: This invention adopts the comprehensive voltage matching distance that integrates the relevant distance and the normalized reconstruction error as the criterion, which makes up for the lack of reliability of a single index under specific working conditions, and makes the final determination result more stable and reliable. Attached Figure Description

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is the overall flowchart of the present invention.

[0024] Figure 2 This is a flowchart of the data acquisition and preprocessing process in this invention.

[0025] Figure 3 This is a flowchart of the user-side photovoltaic impact period identification process in this invention.

[0026] Figure 4 This is a flowchart of voltage-active power response separation and transformer identification in this invention.

[0027] Figure 5 The simulation results are shown in the figure.

[0028] Figure 6 A diagram illustrating the process of identifying the impact periods of photovoltaic self-consumption.

[0029] Figure 7 For voltage Active power response separation result diagram (under actual distribution transformer).

[0030] Figure 8 This is a comprehensive voltage matching distance and discrimination determination diagram.

[0031] Figure 9 A comparison chart showing the accuracy of different identification methods. Detailed Implementation

[0032] The present invention proposes a voltage-based The active power response separation method for identifying the relationship between households and transformers in low-voltage distribution areas completely solves the main technical challenges faced in identifying the relationship between households and transformers in the context of distributed photovoltaic access through data quality governance, photovoltaic self-consumption period identification, response separation modeling, alternating constraint solving, and comprehensive matching judgment. Figure 1 The overall implementation process of the method of the present invention is given, and the steps are described in detail below.

[0033] Based on voltage A method for identifying the relationship between low-voltage transformer substations and their customers based on active power response separation includes the following steps: Step 1: Data Acquisition and Preprocessing This step unifies the measurement data from the user side and the distribution transformer side to provide high-quality input data for subsequent analysis. The specific implementation process is as follows: Figure 2 As shown.

[0034] First, collect data within the target low-voltage distribution area. Voltage time series and active power time series of individual user electricity meters, and Voltage time series at the low-voltage side outlet of each candidate distribution transformer. Due to potential inconsistencies in the time base of the acquisition system, all time series need to be uniformly time-aligned.

[0035] In real-world engineering environments, due to factors such as communication delays, electromagnetic interference, and data acquisition terminal malfunctions, outliers or short-term data gaps inevitably appear in the sampled data. This invention employs a combined approach of exceeding limits criterion and local median absolute deviation criterion to identify outlier data points. For the first... The user in the first Voltage sample value at each sampling time If it deviates from the rated voltage If the deviation exceeds the allowable range, or the deviation from the median within a local time window is greater than the statistical fluctuation threshold, it is judged as an outlier. The specific criteria are as follows: (1); (2); in, To allow for voltage deviation ratio, For time A local time window centered on the center, This indicates taking the median. This represents the absolute deviation of the median of the samples within this window. This is the anomaly detection coefficient.

[0036] For abnormal points in active power measurement, the same logic as for voltage abnormal point identification is used for processing.

[0037] Data points identified as anomalous or originally missing should not be directly deleted, as cutting off the entire curve segment would result in significant information loss. This invention utilizes interpolation to complete the data using neighboring valid sampling points. The completion formula is as follows: (3); In the formula, Generally refers to voltage or active power. The corrected data values, For interpolation operators, linear interpolation or Neville interpolation can be selected based on the characteristics of data variation. and The first The time and value of each of the nearest valid sampling points, It is the set of nearest valid sampling points.

[0038] The reliability of data points from different sources varies. Therefore, this invention assigns different data quality weights to original valid sampling points, interpolation completion points, and outlier correction points. The original valid sampling points are given higher weights, while the interpolated completion points and outlier correction points are given lower weights. In the subsequent model parameter estimation process, these weights make the penalty for the residual contribution of low-confidence data points in the objective function lighter, thereby adaptively reducing the impact of unreliable data on model solution and household change relationship determination.

[0039] For the low-voltage side voltage sequence of candidate distribution transformers, since it is necessary to retain the common voltage difference between each candidate distribution transformer for matching identification, this step only performs anomaly identification and missing information filling, without performing any deduction processing related to photovoltaic disturbances.

[0040] Step 2: Identification of the impact period of user-side photovoltaic self-consumption: After distributed photovoltaic (PV) grid integration, user voltage will show a positive rise relative to the baseline during daytime PV self-consumption periods, while the active power obtained from the grid will decrease accordingly. This coordinated change provides important physical evidence for identifying PV self-consumption periods. For example... Figure 3 As shown, this step utilizes the combined characteristics of voltage rise and active power decrease to pinpoint the moment when the photovoltaic effect is significant.

[0041] First, define the reference voltage baseline. and reference active power baseline The baseline can be obtained by selecting sampled data from the user during non-photovoltaic periods and using methods such as interpolation, smoothing with historical curves of similar weather conditions, or intraday trend filtering to characterize the user's normal voltage and power levels when there is no photovoltaic output. Based on this, the user's... At any moment positive voltage rise and active power decrease : (4); (5); in, and These are the user voltage and active power data after preprocessing in step 1.

[0042] Voltage and power have different dimensions, and directly adding them together cannot reasonably reflect the combined impact of the two characteristics. This invention introduces a scale parameter to standardize the characteristics and constructs a photovoltaic impact score. : (6); In the formula, and These are the scaling parameters for voltage and power changes, respectively, used to eliminate the impact of fluctuation amplitude differences between different users; The voltage rise feature weight is used to adjust the relative importance of the two features, voltage rise and active power decrease, in the scoring. It should be a very small positive number to prevent the denominator from being zero; This is a set of candidate time periods during the day, for example, the time periods during the day when photovoltaic power may be generated.

[0043] During the daytime candidate period, select those that simultaneously meet the following conditions. and The sample was sorted by its score from largest to smallest. ,in This represents the total number of samples that meet the criteria. The more significant the photovoltaic self-consumption characteristic, the higher the score. This invention employs an adaptive threshold strategy, taking the top [samples] after sorting. The percentage score is used as the discrimination threshold. : (7); Then construct the user Moment-time photovoltaic impact indicator matrix : (8); when When, it indicates the user At any moment It exhibits significant photovoltaic self-consumption impact characteristics, and the subsequent response separation model will allow for the retention of a large local positive voltage correction component at this moment; when If the photovoltaic effect is not significant at that time, the local correction will be strongly penalized and tend to be compressed to zero. This time-segmented differentiated penalty strategy not only preserves the effective correction information during photovoltaic periods, but also prevents arbitrary corrections from reducing the fitting error in non-photovoltaic periods or in scenarios with incorrect candidate transformers.

[0044] Step 3: Voltage Active power response separation model construction: The traditional method of relying on voltage curve similarity is ineffective in dealing with photovoltaic disturbances because the user voltage contains multiple physical components, such as the distribution transformer's common voltage, line foundation voltage drop, load current voltage drop, and localized rise in photovoltaic self-consumption voltage. If these components are not separated, photovoltaic differences may be misinterpreted as differences in attribution, leading to misjudgments.

[0045] In view of this, the present invention is applicable to each user and each candidate transformer Establish voltage The active power response separation model separates the pre-processed user voltage. The explicit decomposition is as follows: (9); The model comprises the following components with clearly defined physical meanings: For candidate distribution transformers The pre-processed low-voltage side output voltage represents the common voltage disturbance in the power supply path of the candidate distribution transformer, which is the main source of synchronous voltage changes for users supplied by the same distribution transformer. This is a long-term static voltage bias used to absorb the differences in fixed voltage levels caused by user access location, line length and base load, so that the voltage fluctuations of users at different access locations are comparable after deducting this bias. The equivalent active voltage response coefficient is the ratio of the active power obtained by the user from the grid side. Together they constitute the equivalent active power voltage drop term. It is used to characterize the change in line voltage drop caused by load changes; It is a local positive voltage correction component, specifically used to absorb the non-negative voltage residual caused by distributed photovoltaic self-consumption, and only appears as a non-zero value during the photovoltaic influence period; This is a random error term used to absorb measurement noise, modeling errors, and other unmodeled dynamics.

[0046] To ensure the physical interpretability of the model parameters, reasonable constraints are imposed on each component: (10); in, For the maximum permissible static voltage bias amplitude, This represents the upper limit of the active power response coefficient. The static bias is constrained to a non-positive value, reflecting the actual physical law that the user voltage is usually lower than the distribution transformer outlet voltage; the active power response coefficient is non-negative, ensuring that the voltage drop increases when the load increases; the local positive correction component is non-negative, which is consistent with the physical phenomenon of photovoltaic voltage boosting.

[0047] For any candidate transformer, if it is the actual power supply transformer for the user, then its common voltage... It will have a dominant explanatory power for user voltage changes throughout the entire period, with relatively small residuals except for photovoltaic uplift; conversely, erroneous candidate transformers will not be able to obtain reasonable parameters within the constraints, and the reconstruction error will be significantly larger.

[0048] Step 4: Construction and solution of constrained optimization model: Based on the above model, for each user Candidate pair Construct an optimization model with data quality weights and sparsity penalties, with the objective function as follows: (11); The objective function consists of two terms: the first term is the weighted sum of squared reconstruction errors, where the residuals... The data quality weight is given by equation (9). This reduces the contribution of unreliable sampling points to the objective function; the second term is a local positive correction component. The linear penalty term is designed to suppress unnecessary adjustments and maintain the sparsity and interpretability of the model. Penalty weights Differentiated settings are made based on the photovoltaic impact indication matrix: (12); In the formula, The overall sparsity coefficient controls the overall strength of the penalty term; and These are the disturbance penalty coefficients for the periods of impact from photovoltaic self-consumption and the normal periods, respectively, and are typically set to... This design embodies the strategy of "allowing corrections during photovoltaic periods and suppressing corrections during normal periods." This design enables the model to automatically distinguish between the actual photovoltaic uplift and system biases caused by candidate transformer errors, thus avoiding... The item was abused.

[0049] Because the objective function contains non-smooth terms and data coupling, this invention employs an alternating minimization method for iterative solution. This method divides the parameters into two groups and updates them alternately, optimizing only one group of parameters at a time, ensuring that each subproblem has an efficient solution path. The specific process is as follows.

[0050] set up The current iteration number, initial value .

[0051] Step 4.1: Update the long-term static voltage bias and active response coefficient : Fixed local positive correction term from the previous round Calculate the equivalent voltage difference: (13); The equivalent voltage difference, after deducting the estimated photovoltaic correction from the user voltage, is approximated as being explained solely by static bias and active voltage drop. Based on this, a two-dimensional weighted least squares problem with interval constraints is solved: (14); (15); (16); This subproblem is a simple quadratic programming problem with bounded constraints, which can be solved efficiently.

[0052] Step 4.2: Update the local positive voltage correction component : Fixed update and Calculate the positive residual amount It represents the non-negative margin in the user voltage that is not explained by the model after deducting the distribution transformer common voltage, static bias, and active voltage drop. (17); Then, update using a non-negative soft threshold method. : (18); The physical intuition behind this update is that only when the positive residual exceeds a retention threshold jointly determined by the penalty weights and data quality is the excess portion retained as a local correction component, thus preventing the model from using... To fit random noise or model mismatch.

[0053] Step 4.3: Convergence Judgment: After completing one round of updates, calculate the objective function value. If the relative change between two consecutive iterations is less than a preset threshold ,Right now: (19); Then stop iterating and output the final parameter estimates. , and Otherwise, Then return to step 4.1 and continue execution. In practice, this alternation process usually converges after a few iterations.

[0054] Due to different users The candidate pairings are completely decoupled, allowing this step to be highly parallelized, with a computational complexity of approximately [missing value]. ,in For the number of users, For candidate variables, The total number of sampling points. The average number of iterations is sufficient to meet the engineering calculation needs of medium and large-sized low-voltage distribution areas.

[0055] Step 5: Calculation of overall voltage matching distance: After solving for the parameters of the response separation model, the local photovoltaic correction is subtracted from the user voltage to obtain the perturbation-free user voltage sequence: ; This sequence excludes localized voltage rises caused by photovoltaic self-consumption, better reflecting the underlying dominance of the distribution transformer's common voltage on user voltage. Based on this, the matching degree between candidate distribution transformers and users is evaluated from two dimensions: curve shape and numerical fitting.

[0056] (1) Relevant distance: Calculate the de-disturbance voltage Common voltage with candidate distribution transformer The Pearson correlation coefficient between them was converted into correlation distance: (20); The correlation distance reflects the consistency of the two curves in their temporal trends. Under the condition of actual distribution transformer ownership, the user voltage after deducting photovoltaic correction should be highly synchronized with the common voltage of the distribution transformer, and the correlation distance value should be small.

[0057] (2) Normalized reconstruction error: The weighted reconstruction accuracy of the model over the entire time series is calculated. First, the weighted root mean square voltage reconstruction error is obtained: (twenty one); To eliminate the incomparability caused by differences in voltage fluctuation amplitude among different users, normalization is performed using the standard deviation of the user's original preprocessed voltage: (twenty two); in, The standard deviation operator is used to calculate the standard deviation. The smaller the normalized reconstruction error, the more accurately the candidate distribution transformer model can explain the numerical changes in user voltage.

[0058] (3) Overall voltage matching distance: The two indicators mentioned above measure curve shape consistency and voltage value reconstruction capability, respectively. To balance the advantages of both, they are linearly combined into a comprehensive voltage matching distance: (twenty three); in, The error term weight is used to adjust the relative importance of the two indicators. The smaller the value, the more reasonably the common voltage of the candidate transformer and its equivalent response relationship can explain the complete change process of the user voltage.

[0059] Step 6: Determine the relationship between the household and the property and output the mark to be verified: Obtain all users in step 5 Based on the comprehensive voltage matching distance of candidate distribution transformer pairs, this step ranks the candidate distribution transformers and makes an assignment decision for each user. For example... Figure 4 As shown, For users All The candidate distribution transformers are selected based on the comprehensive voltage matching distance. Sort the variables from smallest to largest, and record the one with the smallest distance as the optimal candidate variable. The one with the smallest distance is the second-best candidate pair. Define the distance difference between the two as the discrimination index: (twenty four); The greater the discrimination, the more obvious the advantage of the optimal candidate pair and the more reliable the attribution relationship.

[0060] The final determination rule uses two conditions: (25); in, This is the maximum allowed matching distance, used to exclude unreliable cases where even the best match has an excessively large distance. To minimize the discrimination threshold and avoid misjudgment due to the close proximity of the optimal and second-best options, a user is assigned to the optimal candidate transformer when both conditions are met. Otherwise, it indicates insufficient confidence in the optimal assignment, which may be due to issues such as incorrect records, multiple transformer boundaries, or poor data quality. The user is then added to the list of candidates to be verified on-site, along with their ranking information, for maintenance personnel to conduct a thorough review based on the actual conditions, such as the location of meter boxes and the routing of low-voltage lines.

[0061] Through the above-described embodiments, the present invention can effectively separate photovoltaic disturbances in low-voltage distribution areas with a high proportion of distributed photovoltaic access, accurately identify the ownership relationship between households and transformers, and provide targeted verification lists for users that are difficult to determine, thus significantly improving the engineering practicality of topology identification.

[0062] Example 1 (I) Simulation Scenarios and Parameter Settings: Building One candidate distribution, A simulation area for low-voltage users, with each transformer supplying power to 4 users, sampling interval... For three consecutive days There are 1 sampling point; among which users , , , 10 including distributed photovoltaic (installed capacity) User voltage is based on the common voltage of the distribution transformer. + Static bias - Active voltage drop + Localized photovoltaic uplift + Random noise The physical mechanism is generated, and outlier points, local outliers, and missing segments of continuous communication are injected to simulate the field data acquisition conditions. The main algorithm parameter values ​​are shown in Table 1: Table 1 Main Algorithm Parameters

[0063] (II) Description of results based on simulation diagrams: Figure 5 Subgraph (a) shows a typical photovoltaic user (user) The original voltage curve of ) The actual voltage curve of the low-voltage side of the power supply transformer shows obvious over-limit spikes (marked with red circles) and continuous missing segments in the original data. Based on the over-limit criterion in S1... Criterion for absolute deviation from local median Joint identification and nearest valid point interpolation completion back, Figure 5 Preprocessed user voltage in subgraph (b) The curve is smooth and complete, eliminating the effects of anomalies and missing values, and also providing a reference voltage baseline obtained from the intraday smoothing trend. It can be seen that S1 assigns differentiated quality weights based on data credibility. The governance process provides highly reliable input for subsequent model estimation.

[0064] Figure 6 Subgraph (a) Figure 6 Subgraph (b) shows the positive voltage rise. With active power decrease Both during the day Significant left-right synchronization was observed, exhibiting a clear synergistic characteristic of voltage rise and power decrease. Figure 6 (c) Score the impact of photovoltaics and before Adaptive threshold determined by proportional position (Equation (7)); Figure 6 (d) refers to the users who obtained the data. Moment-time photovoltaic impact indicator matrix (Equation (8)). It can be seen that... The time periods accurately cover the photovoltaic self-consumption output periods, and the system triggers without error during nighttime and early morning / evening periods, verifying the fine positioning capability of the joint feature and adaptive threshold methods compared to the fixed threshold / entire segment cut-off method.

[0065] Figure 7 In subgraph (a), the de-disturbance voltage after subtracting the local correction component. Common voltage with distribution transformer Highly synchronized, with timing consistency significantly superior to the original user voltage; Figure 7 In subgraph (b), the estimated long-term static voltage bias To satisfy constraints The stable negative value, the equivalent active voltage drop term The voltage drop during the evening peak load period is consistent with the physical law of increased load and increased voltage drop, and the estimated value matches the actual bias and actual response coefficient set in the simulation. Figure 7 In subgraph (c), the estimated local positive voltage correction component It only appears as a non-zero value during photovoltaic periods, and its shape is highly consistent with the actual photovoltaic uplift curve. During ordinary periods... Differentiated sparse penalty weights (Equation (12) satisfies) Under the influence of ), it is compressed to zero. This indicates that the S3 response separation model... The constraints and S4 alternating minimization solution can explicitly decouple photovoltaic perturbations at the mechanistic level. The physical meaning of each component is clear and interpretable, and the convergence condition is met after a small number of iterations. (Equation (19)).

[0066] Figure 8 The subgraph (a) is for users right The comprehensive voltage matching distance of each candidate distribution transformer ,in For relevant distance, Normalized reconstruction error: true distribution transformer (distribution transformer) The smallest distance and the highest discrimination between it and the second-best candidate. Significantly greater than the threshold At the same time satisfy It meets the dual criteria; Figure 8 Subgraph (b) is the matching distance matrix for all 12 users. The true attribution positions all fall at the minimum value of each row, showing a clear block diagonal structure, indicating that the comprehensive distance that combines the relevant distance and the normalized reconstruction error has a stable ability to distinguish the correct attribution.

[0067] Figure 9The diagram compares the accuracy of different identification methods, specifically the direct Pearson correlation method and the method of identifying and eliminating photovoltaic periods. The voltage curves of photovoltaic users are distorted due to localized uplift, making the direct correlation method prone to misclassifying them as non-real transformers with similar voltage patterns. The elimination method, however, suffers from limited accuracy due to the loss of effective information and poor adaptability to users who only consume power during certain periods. The method of this invention explicitly separates the localized photovoltaic uplift component. It has the highest accuracy rate, and under typical operation, all users correctly identify and make no misjudgments. The specific values ​​are subject to the actual output of the program and may vary slightly depending on the random seed and parameter calibration.

Claims

1. Based on voltage The method for identifying the relationship between low-voltage transformer substations and households based on active power response separation is characterized by: Includes the following steps: S1: Obtain the user voltage sequence, user active power sequence and candidate distribution transformer low-voltage side voltage sequence within the target low-voltage distribution area, identify outliers, fill in missing values ​​and assign data quality weights to obtain preprocessed user voltage, user active power and distribution transformer voltage data. S2: Based on the positive rise of user voltage relative to the reference baseline and the decrease of user active power relative to the reference baseline after preprocessing, joint features are constructed to identify the periods of photovoltaic self-consumption impact, forming user... Momentary photovoltaic impact indicator matrix; S3: Build voltage for each user and each candidate transformer. The active power response separation model expresses the preprocessed user voltage as the sum of candidate distribution transformer common voltage, long-term static voltage bias, equivalent active voltage drop, local positive voltage correction component, and random error term; S4: Establish an optimization model with the sum of the weighted reconstruction error and the sparse penalty term of the local positive voltage correction component as the objective function; adopt the alternating minimization method to alternately estimate the long-term static voltage bias, active response coefficient and local positive voltage correction component under the conditions of satisfying the static voltage bias constraint and the active response coefficient constraint. S5: After deducting the local positive voltage correction component, the relevant distance and normalized reconstruction error are calculated based on the user disturbance removal voltage and the candidate distribution transformer common voltage, and then integrated to construct the comprehensive voltage matching distance. S6: For each user, candidate transformers are sorted according to the comprehensive voltage matching distance. The transformer to which the user belongs is determined based on the matching distance of the best candidate transformer and the difference in matching distance between the best and second-best candidate transformers.

2. The voltage-based method according to claim 1 The method for identifying the relationship between low-voltage transformer substations and households based on active power response separation is characterized by: S1 includes the following steps: S1.1: For voltage anomalies, a combined identification method using the over-limit criterion and the local median absolute deviation criterion is employed. For the first... The user in the first Voltage sample value at each sampling time An outlier is identified if any of the following conditions are met: (1); (2); In the formula: For the first The user in the first Voltage sample value at each sampling time. Rated voltage, This is the allowable voltage deviation ratio; The preset convergence threshold, For time A local time window centered on the center, For median operators, The anomaly detection coefficient is used to identify the anomaly. The absolute deviation of the median within a local time window; This represents the voltage sample value of the i-th user within the τ-th local time window; The same method should be used to handle abnormal active power points; S1.2: For short-term missing points or abnormal removal points, interpolation is performed using the nearest valid sampling points to complete the data. The formula is as follows: (3); In the formula: Indicates voltage or active power. This represents the corrected sampled value. This represents the linear interpolation or Neville interpolation operator. and The first The sampling time and sampling value of each neighboring valid sampling point. Represents the set of nearest valid sampling points; S1.3: Based on data reliability, assign different data quality weights to the original valid sampling points, interpolation completion points, and outlier correction points. Among them, the weights of interpolation completion points and outlier correction points are lower than the weights of the original valid sampling points.

3. The voltage-based method according to claim 2 The method for identifying the relationship between low-voltage transformer substations and households based on active power response separation is characterized by, In S1, only outlier identification and missing value completion are performed on the low-voltage side voltage sequence of the candidate distribution transformer, without assigning data quality weights to distinguish the impact of photovoltaics.

4. The voltage-based method according to claim 1 The method for identifying the relationship between low-voltage transformer substations and households based on active power response separation is characterized by: S2 includes: S2.1: Define the reference voltage baseline and reference active power baseline Calculate users At any moment positive voltage rise and active power decrease : (4); (5); In the formula: and users respectively At any moment The reference voltage baseline and reference active power baseline are obtained by interpolation during non-photovoltaic periods, historical curves of similar dates, or intraday smoothing trends; and These are the pre-processed voltage and active power, respectively. S2.2: To unify the dimensions, a scale parameter is introduced to construct the photovoltaic impact score. : (6); In the formula: For users At any moment Photovoltaic impact score, As the voltage rise characteristic weight, This is the voltage rise. For users i The metric for voltage rise. To prevent extremely small positive numbers with a denominator of zero, The decrease in active power, For users i The scale parameter of active power decrease A set of candidate time slots for daytime use; S2.3: Simultaneously satisfying the requirements during the daytime candidate time period and The score of the sample Sort by size from largest to smallest. Take the front Ratio position rating as adaptive threshold : (7); In the formula: For adaptive threshold, The total number of samples participating in the scoring. The proportion of high-scoring samples selected; S2.4: Based on adaptive threshold Constructing users Moment-time photovoltaic impact indicator matrix : (8); In the formula: For users At any moment Photovoltaic impact score, This is an adaptive threshold.

5. The voltage-based method according to claim 4 The method for identifying the relationship between low-voltage transformer substations and households based on active power response separation is characterized by: The reference voltage baseline in S2.1 and reference active power baseline This is obtained by selecting sampled data from the user's non-photovoltaic self-use period and performing interpolation or smoothing.

6. The voltage-based method according to claim 1 The method for identifying the relationship between low-voltage transformer substations and households based on active power response separation is characterized by: Voltage in S3 The active power response separation model and its constraints are as follows: (9); (10); In the formula: This is the pre-processed user voltage; For candidate distribution transformers The pre-processed low-voltage side voltage is used as the candidate distribution transformer common voltage; The long-term static voltage bias to be estimated; The equivalent active voltage response coefficient to be estimated is denoted as . This refers to the preprocessed active power of the user. The local positive voltage correction component to be estimated; This is the random error term; This is the maximum permissible static voltage bias amplitude; This represents the upper limit of the active response coefficient.

7. The voltage-based method according to claim 1 The method for identifying the relationship between low-voltage transformer substations and households based on active power response separation is characterized by: The expressions for the objective function of the optimization model in S4 and the corresponding local positive correction penalty weights are as follows: (11); (12); In the formula: For users In candidate distribution transformers The objective function value is below. For data quality weights, For random error term, The total number of sampling points. The penalty weights are adjusted locally for positive correction. The local positive voltage correction component to be estimated. The overall sparsity coefficient is . To account for the disturbance penalty coefficient during the period of photovoltaic self-consumption, This is the disturbance penalty coefficient corresponding to the normal time period, and it satisfies... , For users Real-time photovoltaic impact indicator matrix.

8. The voltage-based method according to claim 7 The method for identifying the relationship between low-voltage transformer substations and households based on active power response separation is characterized by: The alternating minimization method in S4 includes: S4.1: Let Given the current iteration number, and the local positive voltage correction component from the previous iteration. Calculate the equivalent voltage difference : (13); In the formula: For the first The equivalent voltage difference in the next iteration. For the preprocessed user voltage, For candidate distribution transformers Pre-processing low-voltage side output voltage, This indicates the local correction term from the previous round; S4.2: Solve a two-dimensional weighted least squares problem with interval constraints to update the long-term static voltage bias. and active response coefficient : (14); (15); (16); In the formula: For long-term static voltage bias. The active response coefficient, For the first In the next iteration, regarding b and r The weighted sum of squared residuals, The set of parameter values ​​that satisfy the constraints. b This represents the temporary long-term static voltage bias to be estimated. r This represents the temporary active response coefficient to be estimated. For data quality weights, For the first The equivalent voltage difference in the next iteration. This refers to the preprocessed active power of the user. This is the maximum permissible static voltage bias amplitude; This represents the upper limit of the active power response coefficient. S4.3: Calculate the positive residual based on the updated parameters. The local positive voltage correction component is updated using a non-negative soft threshold method. : (17); (18); In the formula: It is a positive surplus. For the preprocessed user voltage, For candidate distribution transformers Pre-processing low-voltage side output voltage, For long-term static voltage bias. The active response coefficient, This refers to the preprocessed active power of the user. This is a local positive voltage correction component. The penalty weights are adjusted locally for positive correction. For data quality weights, To prevent extremely small positive numbers with a denominator of zero; The retention threshold for the local positive voltage correction component; S4.4: Repeat steps S4.1 to S4.3 until the convergence condition is met. (19); In the formula: and The first Second and third The objective function value of the next iteration. This is the preset convergence threshold.

9. The voltage-based method according to claim 1 The method for identifying the relationship between low-voltage transformer substations and households based on active power response separation is characterized by, S5 includes: S5.1: Calculate the undisturbed voltage sequence Common voltage sequence with candidate distribution transformers The relevant distance between : (20); In the formula: For the relevant distance, The Pearson correlation coefficient; The elements are composed of constitute, For users At any moment Relative to candidate transformers The de-disturbance voltage; This is the pre-processed user voltage; This is a local positive voltage correction component; S5.2: Calculate the normalized reconstruction error, specifically: (21); (22); In the formula: For weighted root mean square voltage reconstruction error, For data quality weights, For random error term, To normalize the voltage reconstruction error, For users The preprocessed voltage sequence, The standard deviation operator; To prevent extremely small positive numbers with a denominator of zero; S5.3: Integrate relevant distances and normalized reconstruction errors to construct a comprehensive voltage matching distance. : (23); In the formula: For comprehensive voltage matching distance, The weights for the error term are used to adjust the contribution ratio of reconstruction error and curve shape distance to the matching distance. To normalize the voltage reconstruction error, For the perturbation voltage sequence Common voltage sequence with candidate distribution transformers The relevant distance between them.

10. The voltage-based method according to claim 1 The method for identifying the relationship between low-voltage transformer substations and households based on active power response separation is characterized by: The process of determining the user's assigned transformer in S6 is as follows: users All candidate distribution transformers are classified according to their comprehensive voltage matching distance. Sort by size from smallest to largest. and Calculate the discrimination index for the best and second-best candidate transformers. : (24); In the formula: For the optimal candidate transformer distance Distance from the second-best candidate transformer difference; The rules for determining user ownership are as follows: (25); In the formula: Indicates user The final identification and attribution of the distribution transformer, This is the optimal candidate transformer. The optimal candidate transformer distance. To the maximum allowed matching distance, This is the minimum discrimination threshold.

Citation Information

Patent Citations

  • Calligraphy guiding method and device, computer equipment and medium

    CN110070089A

  • User-transformer relation identification method based on electric quantity and voltage information fusion

    CN114021430A